Information about how GRE examinees solve verbal analogy problems was obtained in this study through protocol analysis. High- and low-ability subjects who had recently taken the GRE General Test were asked to "think aloud" as they worked through eight analogy items. These items varied factorially on the dimensions of vocabulary difficulty (easy vs. difficult), relationship difficulty (easy vs. difficult), and stem-key correspondence (independent vs. overlapping). A scoring system was developed that included three phases: (i) a cognitive process analysis, (ii) a global description of the subjects' strategies, and (iii) an evaluation of their performance. Overall we found that vocabulary and correspondence tended to have multiple effects on how subjects solved analogies and affected whether they solved them correctly. On the other hand, relationship difficulty tended to affect whether or not subjects achieved correct solutions but not how they attained solutions. As expected, high-ability subjects tended to use higher-level or more sophisticated strategies more often than low-ability subjects. However, the use of higher-level strategies was common for both groups of subjects. The results are compared with those reported in the cognitive science literature. In addition, the implication of these results for test development and the usefulness of the protocol methodology are discussed.